This thesis implemented a field programmable gate array (FPGA)-based face detector using a neural network (NN), as well as a bit-width reduced floating-point unit (FPU). An NN was used to easily separate face data and non-face data in the face detector. The NN performs time consuming repetitive calculation. This time consuming problem was solved by a Field Programmable Gate Array (FPGA) device and a bit-width reduced FPU in this thesis. A floating-point bit-width reduction provided a significant saving of hardware resources, such as area and power.<p>The analytical error model, using the maximum relative representation error (MRRE) and the average relative representation error (ARRE), was developed to obtain the maximum and average output errors for the bit-width reduced FPUs. After the development of the analytical error model, the bit-width reduced FPUs and an NN were designed using MATLAB and VHDL. Finally, the analytical (MATLAB) results, along with the experimental (VHDL) results, were compared. The analytical results and the experimental results showed conformity of shape. It was also found that while maintaining 94.1% detection accuracy, a reduction in bit-width from 32 bits to 16 bits reduced the size of memory and arithmetic units by 50%, and the total power consumption by 14.7%.

Die Modellierung natürlicher Neuronenpopulationen stellt den Versuch dar, komplizierte Wechselwirkungen und Ereignisabhängigkeiten in biologischen Systemen quantitativ erfassen zu wollen. Widersprüchlich erscheint dabei die Tatsache, daß ein einzelnes Neuron in einer Population ohne Signifikanz ist, daß sich die gesamte Population aber aus einer Vielzahl derartiger Neuronen zusammensetzt und eine, technischen Systemen überlegene funktionelle Vielfalt besitzt /ZUR 92/, /HOL 93/. Ergo setzt sich die Gesamtleistung eines Systems nicht aus der Summe der Leistungen seiner Komponenten "summarisch" zusammen, sondern resultiert vielmehr aus deren Wechselwirkungen. Technisch interessant erscheinen an dieser Stelle mindestens zwei Fragen: * Welcher Mechanismus begründet den genannten Widerspruch in Neuronenpopulationen? * Welche technische Anleihe bietet dieser Mechanismus? Die Modellierung einer Neuronenpopulation kann auf zweierlei Art und Weise erfolgen. Entweder durch die Aufklärung der Neuronenpopulation "von innen heraus", d.h. durch Beobachtung und mathematische Formulierung physiologischer Abläufe oder durch vergleichende Betrachtungen mit "konvergenten" Modellen, d.h. durch die Schaffung von Modellen mit vergleichbaren Phänomenen. Die nachfolgenden Ausführungen favorisieren die letztgenannte Vorgehensweise. Phänomene sind die ereignisabhängigen Schwellwertentwicklungen der Neuronen in Wechselwirkung mit den umgebenden Neuronen sowie die ereignisabhängigen Entwicklungen der synaptischen Verbindungsstärken zwischen den Neuronen, bezeichnet als "Leitwertentwicklung". Technische Anwendungen dieser Simulationsergebnisse werden erörtert, zum Beispiel die Nachbildung der Durchdringung diffusionsfähiger Medien mit Schadstoffen und die Objektvereinzelung.

Sleep staging can be used to assess whether sleep structure is abnormal. According to the R&K rule, human sleep can be divided into four different stages: Awake, Light Sleep, Deep Sleep and Rapid-Eye-Movement (REM) Sleep. Conventionally, sleep staging are scored mainly by EEG signals and complementally by EOG and EMG signals. The goal of this study is to detect slow wave sleep (SWS) automatically by using different combinations of EEG, EOG, and EMG signals. In particular, a total of 16 combinations of channels have been studied. Based on high amplitude slow wave characteristics of SWS, this study develops many of feature variables to characterize SWS. A subset of these features are employed to design neural network classifier to detect SWS. This study has noted interpersonal-differences in physiological signals between people and proposes solutions to this problem to improve the performance of SWS detection. The number of tested subjects from two different sleep centers is 1318 and 947 subjects, respectively. These subjects were divided into five groups for training and testing data in order to test performance of our proposed approach. By applying the proposed approach to 1318 subjects, the experimental results show that the proposed method achieves kappa of 0.63 by using a single EEG channel, kappa of 0.6 by using two channels EOG and kappa of 0.66 by using the best combination of multi-channel singals. The size of dataset used in this work is significantly large than those of previous studies and thus provide more reliable experimental results. The experimental results show that the proposed approach can provide satisfactory performance in dealing with dataset with more than 1000 subjects.

An isolated hybrid system comprised of a dispatchable and a non-dispatchable power generation sources, is proposed to supply the load of a remote village in the west coast region of The Gambia. The thesis presents an artificial neural network (ANN) based approach to tune the parameters of the frequency regulator in hybrid wind/diesel power system for isolated area power supply. The multi-layer feed-forward ANN with the error back-propagation training is employed to tune the frequency regulator in the simulation of hybrid system under different load and wind conditions. Using MATLAB/Simulink, dynamic simulations are performed to investigate the interaction between these two power sources for the load management, and the voltage and frequency behaviors during wind speed and load variations. Simulation results show that the wind turbine and the diesel generator can be operated suitably in parallel. During simulation, the frequency and voltage regulators used in the proposed hybrid system performed fairly well under wind speed variations and load changing conditions. A good frequency regulator interface, which is around 50Hz is observed for nearly the entire period of operation.

The wave power converstion system nowadays nearly all have to depend on the converstion of mechanical energy.This way frequently causes unnecessary loss to the power.The costs of maintenance will also be high about the way. Therefore,we bring up a new wave power converstion system according to Faraday's Law.This way is no need to have the turbine.It can be catch the induced current from the generator directly. We use three different types of the PVC tubes as the model of the magnetic field and put into the round magnet.The motion of the round magnet will cause the change of the magnetic field to product the induced current.We install the magnet with the tube on a platform and combine them to be a structure.Different tubes, structure period and structure displasement will cause different effect of the generator.We apply to the data of experiments to find the relationship of the generator, structure period and structure displacement.We also use the neural network to build the model of the relationship. Finally, it will be the basis on the design of the real model in the future.

The artificial neural network (ANN) approach has been recognized as a capable technique to model the highly complex and nonlinear problem of travel time prediction. In addition to the nonlinearity, a traffic system is also temporally and spatially dynamic. Addressing the temporal-spatial relationships of a traffic system in the context of neural networks, however, has not received much attention. Furthermore, many of the past studies have not fully explored the inclusion of incident information into the ANN model development, despite that incident might be a major source of prediction degradations. Additionally, directly deriving corridor travel times in a one-step manner raises some intractable problems, such as pairing input-target data, which have not yet been adequately discussed. In this study, the corridor travel time prediction problem has been divided into two stages with the first stage on prediction of the segment travel time and the second stage on corridor travel time aggregation methodologies of the predicted segmental results. To address the dynamic nature of traffic system that are often under the influence of incidents, time delay neural network (TDNN), state-space neural network (SSNN), and an extended state-space neural network (ExtSSNN) that incorporates incident inputs are evaluated for travel time prediction along with a traditional back propagation neural network (BP) and compared with baseline methods based on historical data. In the first stage, the empirical results show that the SSNN and ExtSSNN, which are both trained with Bayesian regulated Levenberg Marquardt algorithm, outperform other models. It is also concluded that the incident information is redundant to the travel time prediction problem with speed and volume data as inputs. In the second stage, the evaluations on the applications of the SSNN model to predict snapshot travel times and experienced travel times are made. The outcomes of these evaluations are satisfactory and the method is found to be practically significant in that it (1) explicitly reconstructs the temporalspatial traffic dynamics in the model, (2) is extendable to arbitrary O-D pairs without complete retraining of the model, and (3) can be used to predict both traveler experiences and system overall conditions.

Reflection cracking is one of the main distresses in hot-mix asphalt (HMA) overlays. It has been a serious concern since early in the 20th century. Since then, several models have been developed to predict the extent and severity of reflection cracking in HMA overlays. However, only limited research has been performed to evaluate and calibrate these models. In this dissertation, mechanistic-based models are calibrated to field data of over 400 overlay test sections to produce a design process for predicting reflection cracks. Three cracking mechanisms: bending, shearing traffic stresses, and thermal stress are taken into account to evaluate the rate of growth of the three increasing levels of distress severity: low, medium, and high. The cumulative damage done by all three cracking mechanisms is used to predict the number of days for the reflection crack to reach the surface of the overlay. The result of this calculation is calibrated to the observed field data (severity and extent) which has been fitted with an S-shaped curve. In the mechanistic computations, material properties and fracture-related stress intensity factors are generated using efficient Artificial Neural Network (ANN) algorithms. In the bending and shearing traffic stress models, the traffic was represented by axle load spectra. In the thermal stress model, a recently developed temperature model was used to predict the temperature at the crack tips. This process was developed to analyze various overlay structures. HMA overlays over either asphalt pavement or jointed concrete pavement in all four major climatic zones are discussed in this dissertation. The results of this calculated mechanistic approach showed its ability to efficiently reproduce field observations of the growth, extent, and severity of reflection cracking. The most important contribution to crack growth was found to be thermal stress. The computer running time for a twenty-year prediction of a typical overlay was between one and four minutes.

The technology of power electronics is used increasingly during recent years, and the electronic power facilities are used more and more in the power system. The non-linear electronic loads produce heavy harmonic currents and could significantly degrade the power quality. Nonlinear loads, including the un-interruptible power supply, motor control and converter, etc, are important equipment in a modern factory, however, these nonlinear loads could lead to power facility malfunction and capacitor damage. The harmonics would eventually cause severe unexpected capital loss. Power quality has become an important study. This thesis proposes the probabilistic neural network (PNN) for power harmonics detection from distorted waves. Originally, Fourier transform is often used to analyze distorted waves in frequency spectrum, and low-pass filter is used to eliminate the fundamental component where characteristic harmonics can be detected. The complicated process is difficult to operate in real time. PNN based processing model with physical harmonic data is used to simplify the process. Computer simulation will show a simplified model and shorter processing time for harmonic detection in the active filter. The Intranet based distributed characteristic harmonic monitoring system.LabVIEW language was used to develop the Human-Machine Interface(HMI) , and DataSocket tool was used to share the information on net.

Co-generation is an efficient energy system that generates steam and electricity simultaneously. In ordinary operation, fuel cost accounts for more than 60% of the operational cost. As a result, the boiler efficiency and optimization level of co-generation are both high. To achieve further energy conservation, objectives of this thesis are to find the Profit-maximizing dispatch and efficiency enhancing strategy of the co-generation systems under deregulation. In a coexistent environment of both Bilateral and Poolco-based power market, there are bid-based spot dispatch, and purchases and sales agreement-based contract dispatch. For profit-maximizing dispatch, the steam of boilers, fuels and generation output will be obtained by using the SQP(Sequential Quadratic Programming ) method. In order to improve the boiler efficiency, this thesis utilizes artificial neural networks(ANN) and evolutionary programming(EP) methods to search for the optimal operating conditions of boilers. A co-generation system (back-pressure type and extraction type) is used to illustrate the effectiveness of the proposed method.

Taguchi methods is also called quality engineering. It is a systematic methodology for product design(modify) and process design(improvement) with the most of saving cost and time, in order to satisfy customer requirement. Taguchi¡¦s parameter design is also known as robust design, which has the merits of low cost and high efficiency, and can achieve the activities of product quality design, management and improvement, consequently to reinforce the competitive ability of business. It is a worthy research course to study how to effectively apply parameter design, to shorten time spending on research, early to promote product having low cost and high quality on sale and to reinforce competitive advantage. However, the parameter design optimization problems are difficult in practical application owing to (1)complex and nonlinear relationships exist among the system¡¦s inputs, outputs and parameters and (2)interactions may occur among parameters. (3)In Taguchi¡¦s two-phase optimization procedure, the adjustment factor cannot be guaranteed to exist in practice. (4)For some reasons, the data may become lost or were never available. For these incomplete data, the Taguchi¡¦s method cannot treat them well. Neural networks have learning capacity fault tolerance and model-free characteristics. These characteristics support the neural networks as a competitive tool in processing multivariable input-output implementation. The successful field including diagnostics, robotics, scheduling, decision-marking, predicition, etc. In the process of searching optimization, genetic algorithm can avoid local optimization. So that it may enhance the possibility of global optimization. This study had drawn out the key parameters from the spheroidizing theory, and L18, L9 orthogonal experimental array were applied to determine the optimal operation parameters by Signal/Noise analysis. The conclusions are summarized as follows: 1. The spheroidizing of AISI 3130 used to be the highest unqualified product, and required for the second annealing treatment. The operational record before improvement showed 83 tons of the 3130 steel were required for the second treatment. The optimal operation parameters had been defined by L18(61¡Ñ35) orthogonal experimental array. The control parameters of the annealing temperature was at B2